captures
Three runs of the pinned build, one per browser, taken on 2026-08-07 on the same macOS machine through npm run capture.
Three runs of the pinned build, one per browser, taken on 2026-08-07 on the same macOS machine
through npm run capture.
| file | browser | requests | signal ids on the wire |
|---|---|---|---|
chrome-2026-08-07T16-01-51-717Z.json | Chrome 149 | 2 | 144 |
firefox-2026-08-07T15-59-37-104Z.json | Firefox 151 | 2 | 144 |
safari-2026-08-07T16-00-02-124Z.json | Safari 26.4.1 | 16 | 144 |
Safari's sixteen requests are the retry queue: the demo tenant rejects a localhost origin, so the
agent posted the same frame seven times and re-fetched the GET leg between attempts.
A capture holds the requests with their frame bytes (reqRaw), the in-page decode, the API call
log, the worker messages, the probe counts and a storage snapshot. Third-party storage entries are
dropped on save, and npm run scrub does the same to a file taken earlier.
The frames in the files here were rebuilt rather than kept as sent: npm run scrub -- --reseal
re-encodes each request's decoded payload with the agent's own codec and writes the result back to
reqRaw, so opening a frame gives the same values the file shows in plain text. The frame layout,
the compression threshold and the seal are the agent's, so node tools/codec.mjs --open reads them
like any other frame, but the byte-for-byte original request is not published.
npm run join reads every file here and writes reference/observed.md. npm run replay and
npm run profile take one as a donor.